Lecture 13A: Evolutionary Robotics

Question 1

What is one origin of evolutionary robotics?

Answer 1

Brooks’ behaviour-based robotics, which divides the system based on level of behaviours.

Bio-inspired approach follows the idea of natural selection for behaviours.

Question 2

What are the 5 stages of the Genetic Algorithm?

Answer 2

Initialisation -> (Fitness test -> Selection -> Reproduction -> Mutation)

• the part in brackets loops

Question 3

What is an example of evolutionary robotics?

Answer 3

Khepera’s body and brain has neural networks with evolving weights.

FARSA simulator is an integrated open-source C++ libraries for evolutionary experiments with robots. It is used in iCub.

Question 4

Explain co-evolution and arm race using an example.

Answer 4

Arm race in evolutionary robotics can result in situations where progress in one population causes the complexification of the task due to the parallel progress in the competing population. Each population gets gradually better than the other.

Co-evolution with Kheperas is seen when the performance of one agent affects the performance of the other agents, and vice versa. It selects the most effective solutions to match counter-strategies. IComplexity without increased supervision

Question 5

What are evolving morphologies? Give an example.

Answer 5

Evolving morphologies convert direct genotype-phenotype encoding to grammar encoding.

Examples:
Kitano (evolving network’s connections)
Sims (evolving virtual creatures)
Bongard & Lipson (evolving robots)

Question 6

Why is evolutionary robotics useful?

Answer 6

The combination of evolution and learning enables robots to adapt to their changing environment.

Evolutionary algorithms can model ontogenetic learning.

The evolutionary method can be applied to encourage cooperative behaviour, especially when robots are genetically related.

Evolutionary robotics provides an ideal framework for studying the role of sensory-motor coordination in developing cognitive and behavioural skills.